House investigation into DeepSeek teases out funding, security realities around Chinese AI tool

A House panel has concluded that the U.S. government should double down on export controls and other tools to slow down the progress of Chinese AI companies like DeepSeek, while also preparing for a future where those efforts fail.

In a report released Wednesday, the House Select Committee on the Chinese Communist Party further fleshes out the financial and technological resources that went into building DeepSeek’s R1 reasoning model, as well as its potential risks to U.S. economic and national security.

The authors conclude that the DeepSeek website and app “acts as a direct channel for foreign intelligence gathering on Americans’ private data.”

“With its direct ties to China’s security and surveillance infrastructure and its unchecked data collection practices, it can function as an open-source intelligence asset feeding American user data into an adversarial system,” the report claims.

DeepSeek’s R1 release caused a panic in U.S. markets earlier this year and signaled that Chinese companies were much further along in the AI race than initially assumed. The release also called into question whether the capital and compute-intensive approach taken by U.S. AI companies to build their own commercial models was warranted, given that R1 requires significantly less processing power for similar or better performance.

One of the most shocking contentions was that R1 had cost just $6 million to build. The fact that it was allegedly built as the side project of a quantitative financial firm with little experience in the AI field caused even more alarm.

While these claims were always heavily qualified — the $6 million figure cited was only the cost of R1’s final training run, leaving out the much larger foundational training and costs needed to get to that point — the House report makes it clear that DeepSeek was far from a scrappy startup.

While DeepSeek is officially owned by the much larger Ningbo Cheng’en Enterprise Management Consulting Partnership, it is effectively controlled by Liang Wenfeng, who owns a majority stake in Ningbo Cheng’en.

Wenfeng is also the head of High-Flyer Quant, a Chinese quantitative trading firm that pumped at least $420 million in initial investment funding into DeepSeek, while also opening up access to the firm’s coveted Firefly supercomputing infrastructure, with 10,000 NVIDIA A100 GPUs. According to analytics company SemiAnalysis, DeepSeek likely has access to 60,000 NVIDIA chips.

While High-Flyer Quant and DeepSeek are technically separate entities, their overlapping ownership and IT infrastructure function as “an integrated ecosystem under Liang’s control,” the report contends.

High-Flyer Quant, as AI experts have pointed out, is “obsessed with their computing infrastructure,” which was already designed to chase “nanosecond advantages in beating the market.” That made Firefly a powerful engine that was already optimized for training AI models.

DeepSeek is also part of the Hangzhou Chengxi Science and Technology Innovation Corridor, a Chinese state-sponsored economic development agency that seeks to support the creation of China’s own Silicon Valley, and has ties to other strategic state-linked hardware companies like Zhejiang Lab.

The report also endorses several other claims that have surfaced around DeepSeek’s cyber and data security in recent months, namely that the app illegally distilled training data from U.S. models, circumvented export controls around computer chips and funnels detailed user data to China.

The House committee held meetings with U.S. industry leaders, who have argued there is evidence that DeepSeek employees used aliases and international banking channels to purchase “dozens” of accounts for leading U.S. AI models. They allegedly used that access to copy the work of U.S. models and create their own imitation.

An unnamed OpenAI executive is quoted in a letter to the committee, claiming that an internal review found that “DeepSeek employees circumvented guardrails in OpenAI’s models to extract reasoning outputs, which can be used in a technique known as ‘distillation’ to accelerate the development of advanced model reasoning capabilities at a lower cost.”

The claim is ironic, as OpenAI and other U.S. companies are subject to numerous domestic lawsuits that allege they similarly appropriated massive amounts of protected intellectual property while training their own algorithms.

DeepSeek also censors anti-CCP sentiments and collects and sends user data, including chat history, device details and typing cadence, via backend infrastructure to China Mobile — a telecom that has been designated by the U.S. as a Chinese military company. Like many domestic U.S. social media apps, DeepSeek is also riddled with tracking tools and other methods for extracting granular data about its users.

This creates a similar potential threat posed by other Chinese companies like Huawei and ZTE, who are also subject to national security laws that require handing over user data to the Chinese government upon request. DeepSeek also does not encrypt much of the data it transmits, making it vulnerable to interception by malicious hackers.

“While the extent of data transmission remains unconfirmed, DeepSeek’s integration with China Mobile infrastructure raises serious concerns about potential foreign access to Americans’ private information,” the report states.

The report recommends that U.S. policymakers tighten export controls on semiconductor chips and other relevant AI technologies, expand funding and resourcing for the Commerce Department’s export controls office and prohibit federal agencies from purchasing or using Chinese AI models.

However, the authors also warn that the U.S. should prepare for a future where current efforts to constrain international rivals like China from building high-quality AI systems falter. That should include empowering Commerce and other departments to more proactively track companies like DeepSeek and its strategic providers in their more nascent stages, allowing for faster policy responses from Washington.

“The emergence of DeepSeek is a warning to U.S. policy makers that the PRC remains capable of rapidly innovating in today’s most advanced technologies despite U.S. efforts to stop them,” the authors wrote. “As AI continues to advance in capability, U.S. departments and agencies must improve their piecemeal approach to prevent strategic surprises that may prove destabilizing.”